Agent Loop Classification

This document defines which agents are part of the Inner Loop vs Outer Loop in the Spec-Driven Development workflow.

Overview

The development workflow is divided into two loops:

  • Inner Loop: Fast, local iteration cycle (edit → run/tests → debug → repeat) focused on instant feedback and developer flow
  • Outer Loop: Post-commit CI/CD pipeline (PR → build → test → package → attest → deploy → observe) focused on consistency, safety, and production reliability

Inner Loop Agents

The inner loop is optimized for developer velocity and rapid iteration. These agents support local development, testing, and validation before commit.

Planning Agents

These agents help plan and design features before implementation:

Agent Description Primary Use
product-requirements-manager PM/PRD creation and management Create detailed product requirements and user stories
software-architect System architecture and design patterns Design system architecture, ADRs, and integration patterns
platform-engineer Platform and infrastructure planning Design CI/CD, infrastructure, and platform architecture
researcher Research and analysis Conduct research for technical decisions and validation
business-validator Business case validation Validate business value and feasibility

Implementation Agents

These agents write production code:

Agent Description Primary Use
frontend-engineer React/React Native development Build UI components, state management, and mobile apps
backend-engineer API and services development Build REST/GraphQL/gRPC APIs, business logic, and data models
ai-ml-engineer ML/AI implementation Build ML pipelines, models, and inference services

Code Review Agents

These agents review code for quality, security, and best practices:

Agent Description Primary Use
frontend-code-reviewer Frontend code review Review React/mobile code for performance, accessibility, and quality
backend-code-reviewer Backend code review Review API/service code for security, performance, and architecture
python-code-reviewer Python code review Review Python code for best practices and quality

Testing & Quality Agents

These agents ensure quality through testing:

Agent Description Primary Use
quality-guardian Comprehensive QA testing Functional, integration, performance, and non-functional testing
playwright-test-generator E2E test generation Generate Playwright tests for user workflows
playwright-test-healer Test maintenance Fix broken tests and improve test reliability
playwright-test-planner Test planning Plan comprehensive test coverage strategy

Security Agents

These agents ensure security throughout development:

Agent Description Primary Use
secure-by-design-engineer Security validation Code security review, dependency scanning, threat modeling

Language Specialist Agents

These agents provide language-specific expertise:

Agent Description Primary Use
go-expert-developer Go development Go-specific implementation and best practices
java-developer Java development Java-specific implementation and best practices
js-ts-expert-developer JavaScript/TypeScript JS/TS-specific implementation and patterns
go-expert-advisor Go advisory Go architecture and design guidance

Documentation Agents

These agents create and maintain documentation:

Agent Description Primary Use
tech-writer Technical documentation API docs, user guides, release notes, technical documentation
star-framework-writer STAR framework writing Create structured narrative documentation

Other Agents

Agent Description Primary Use
executive-tech-recruiter Technical recruiting Assist with technical hiring and job descriptions

Outer Loop Agents

The outer loop ensures organizational safety, compliance, and production reliability. These agents handle release coordination and operational concerns after commit.

Agent Description Primary Use
release-manager Release and deployment management Release readiness assessment, deployment strategy, promotion workflows, change management, production releases

Note: sre-agent and /flow:operate have been removed from the standard workflow. For SRE operations, use /ops:* commands directly (/ops:monitor, /ops:respond, /ops:scale).

Inner Loop Requirements

As defined in task-004.md, the inner loop MUST include:

Testing & CI Simulation

  • Local testing capabilities with options to build as much of CI as possible before commit
  • Fast feedback loops (< 2s hot reload)
  • Local CI execution matching remote CI to catch issues before push
  • Pre-commit validation with hooks and local checks

Validation & Contracts

  • Validation tests/contracts for specific key boundaries and interfaces
  • API contract testing to ensure compatibility
  • Integration point validation between components
  • Fast unit tests with watch mode

Development Environment

  • Containerized dev environment with production parity
  • Hot reload capabilities for immediate feedback
  • Local mocks and emulators for external dependencies
  • Automated dependency installation

Outer Loop Requirements

As defined in task-004.md, the outer loop MUST include:

CI/CD Pipeline (GitHub Actions)

  • Complete GitHub Actions process for CI to:
    • Build artifacts
    • Run comprehensive test suites
    • Scan for security vulnerabilities (SAST, DAST, SCA)
    • Generate SBOM (Software Bill of Materials)
    • Sign artifacts cryptographically

Continuous Delivery

  • CD step to promote artifacts upon:
    • Testing completion and validation
    • Required approvals (manual/automated)
    • Security scan passage
    • Quality gate satisfaction

Stack-Specific CI

  • Stack-specific CI steps that:
    • Work with the specific technology stack in use
    • Don't overpopulate .github/workflows/ folder by default
    • Are only added once the stack is validated through the correct process (before implementation)

Build-Once, Promote Everywhere

  • Build exactly once in dev/CI environment
  • Promote immutable artifacts to staging and production
  • NEVER rebuild for different environments
  • Environment-agnostic artifacts with externalized configuration

Security & Compliance

  • Supply chain security with SLSA compliance
  • Artifact signing and provenance attestation
  • Policy-as-code enforcement
  • Complete audit trails for all pipeline activities

Agent Usage in Workflows

/flow:specify (Inner Loop)

Uses: product-requirements-manager

/flow:plan (Inner Loop)

Uses: software-architect-enhanced, platform-engineer-enhanced

/flow:research (Inner Loop)

Uses: researcher, business-validator

/flow:implement (Inner Loop)

Uses: frontend-engineer, backend-engineer, ai-ml-engineer, frontend-code-reviewer, backend-code-reviewer

/flow:validate (Inner Loop with Outer Loop preparation)

Uses: quality-guardian, secure-by-design-engineer, tech-writer, release-manager

Note: The validate workflow bridges inner and outer loops by preparing artifacts for outer loop deployment while still part of inner loop quality checks.

/flow:operate (REMOVED)

Note: /flow:operate has been removed. Use /ops:* commands directly for SRE operations.

Best Practices

Inner Loop

  1. Keep iterations fast: Every tool and check should run in seconds, not minutes
  2. Test locally first: Validate changes locally before pushing to CI
  3. Use pre-commit hooks: Catch issues before they enter version control
  4. Mirror CI locally: Run the same checks locally that will run in CI
  5. Fail fast: Surface errors immediately for quick resolution

Outer Loop

  1. Automate everything: Manual steps are error-prone and slow
  2. Build once, deploy many: Never rebuild for different environments
  3. Sign all artifacts: Ensure cryptographic verification of all releases
  4. Progressive delivery: Use canary/blue-green for safe deployments
  5. Monitor continuously: Observe system health and performance metrics
  6. Practice rollback: Regular rollback drills ensure preparedness

Loop Transition Points

From Inner to Outer Loop

The transition happens at commit/push to main branch:

  1. Developer completes inner loop:

    • Planning, implementation, local testing complete
    • Pre-commit hooks pass
    • Local CI simulation passes
    • Code reviewed and approved
  2. Commit triggers outer loop:

    • PR created and merged
    • CI pipeline executes (build, test, scan, sign)
    • Artifacts stored in registry
    • Deployment pipelines triggered

From Outer to Inner Loop

Feedback flows back from outer loop to inform inner loop:

  1. Production insights:

    • Monitoring and observability data
    • Performance metrics and issues
    • User feedback and bugs
  2. Inner loop adjustments:

    • Update tests based on production issues
    • Refine monitoring and alerting
    • Improve error handling
    • Enhance documentation

References